Unexpected behaviour of tl431 based regulator

[spec]

Spec did you ever cast an eye over this circuit again? I'm not in a hurry to get it up and running, it was just an idea. Perhaps it could be used as a linear post regulator along with some of my / our SMPS circuits.

Sorry Gordon, I was all talk and no trousers on this one. I have looked it over though and have the information in my head. I will post quite soon and describe what I think. A linear post regulator is a good idea.

It is 1:50am. Why aren't you in bed asleep? Me, I can't stop coughing.

 

[Gasboss775]

Stayed up late watching a film, just checking in here briefly before sleep ( hopefully ) Thanking you in advance for any help with the tl431 circuit.

 

[spec]

TL431 data sheet:
https://www.ti.com/lit/ds/symlink/tl431.pdf
2N5458 data sheet:
https://www.fairchildsemi.com/datasheets/2N/2N5458.pdf
BD139 data sheet:
https://www.fairchildsemi.com/datasheets/BD/BD135.pdf

Gordon, the backing data for your #1 PSU is given above. I have redrawn your circuit so that it can be more conveniently referred to in this post, but the circuit is exactly as your original schematic, except FET type number corrected.

As I have said before, this is a clever design and I understand the objectives:
(1) Use minimum components
(2) use a JFET as a constant current generator to both isolate any ripple/noise on the input voltage from the voltage control functions and also to provide a high gain and therefore high accuracy for the voltage control feedback loop.
(3) Use a cheap but non-the-less accurate voltage reference and amplifier (TL431)

Unfortunately the circuit has a few practical issues:

(1) There is so much voltage gain that it would be very difficult to stabilise it in the frequency domain. This applies not only to the phase changes in the main amplifying path but also the various signal paths due mainly to parasitic capacitances.

(2) The constant current generated by the JFET will have a very wide range due to the wide Idss (vgs=0) of 2ma to 9ma.

(3) The JFET Id saturation current starts at 2V. At a Vds less than this the JFET acts, not as a constant current generator, but as a resistance.

(4) For some reason, this type of constant current generator is never very happy and is inclined to be temperamental. A far better performing, but less elegant, constant current generator can be made with a couple of bipolar junction transistors.

(5) The TL431 is a shunt regulator so it cannot source current to drive the base of the series pass power bipolar transistor. Thus the current for the base is the constant current passed by the JFET less 1ma required for the TL431 to operate. Depending on the particular JFET used this would give a base current of 1 to 8 mA. The Hfe (current gain) of the BD139 is 25 so the maximum current the BD139 could output would only be 25mA to 200mA. Because of other factors the output current would be less than this.

(6) Even if a resistor replaced the JFET, this is still a very high gain circuit. The TL431 needs a high gain to achieve its accuracy. This means that frequency tailoring will be required to ensure frequency stability. You can see evidence of this in the TL431 data sheet in the applications section which shows a PSU similar to your #1 PSU but without the JFET constant current generator. The capacitor shown shapes the overall frequency response to ensure stability. This is standard practice for high gain feedback amplifiers. Our old friend the uA741 op amp has a massive 30p capacitor fabricated into the circuit for this sole purpose.

(7) Finally a general point. There is no decoupling on PSU#1. Just to give you a clue, if you look at a well-designed piece of analogue commercial equipment you will often find capacitors splattered all over the place. This is especially so with high end audiophile amplifiers. The reason for this is that to the amplifier the supply lines and other critical points must look like OV (ground) to them, especially as the frequency increases. If that is not the case the physical circuit bears no resemblance to schematic that you work to. This is a big and complicated subject but for the average circuit you can get by with a few basic rules. There is a standard saying with analogue designers: ‘hundred nan to deck’. This means decouple all critical areas with a 100nF low loss capacitor (normally ceramic) to ground or OV.

That takes care of the high frequencies in most cases, but you also need to decouple at low frequencies too, normally with a 47uF upwards low loss aluminum electrolytic or tantalum capacitor.

Decoupling is not just a thing done by some pedantic analogue design engineers, it is essential. You are probably thinking that you have seen hundreds of circuits with no decoupling. And so you will. But my experience is that many home constructors come to grief because of frequency instability cause by insufficient loop frequency tailoring, no decoupling and poor layout. Decoupling is not always necessary, but it is always a wise precaution. Take the uA741 for example. It is a very friendly opamp and is slugged down to have a narrow frequency response so will tolerate almost anything, but it will still perform best with a good layout and decoupling.

So in summary Gordon you innovative design has been let down by the shortcomings of practical components- they are just not up to your high expectations. I’m sorry to say this but for the reasons already outlined and other reasons which I haven’t gone into, this circuit would be difficult to tame. But no worries, I will post a circuit that will make up for my negative view and give you something to really get your teeth into.

 

[Gasboss775]

TL431 data sheet:
https://www.ti.com/lit/ds/symlink/tl431.pdf
2N5458 data sheet:
https://www.fairchildsemi.com/datasheets/2N/2N5458.pdf
BD139 data sheet:
https://www.fairchildsemi.com/datasheets/BD/BD135.pdf

View attachment 96858​

Gordon, the backing data for your #1 PSU is given above. I have redrawn your circuit so that it can be more conveniently referred to in this post, but the circuit is exactly as your original schematic, except FET type number corrected.

As I have said before, this is a clever design and I understand the objectives:
(1) Use minimum components
(2) use a JFET as a constant current generator to both isolate any ripple/noise on the input voltage from the voltage control functions and also to provide a high gain and therefore high accuracy for the voltage control feedback loop.
(3) Use a cheap but non-the-less accurate voltage reference and amplifier (TL431)

Unfortunately the circuit has a few practical issues:

(1) There is so much voltage gain that it would be very difficult to stabilise it in the frequency domain. This applies not only to the phase changes in the main amplifying path but also the various signal paths due mainly to parasitic capacitances.

Yes, the circuit was assembled on a breadboard, I'd imagine there would have been a fair bit of stray capacitance due to the breadboard.

(2) The constant current generated by the JFET will have a very wide range due to the wide Idss (vgs=0) of 2ma to 9ma.

As a one of design I had the liberty of premeasuring the Idss of the jfet, I think it was between 6-7 mA. That was lucky though, I guess.

(3) The JFET Id saturation current starts at 2V. At a Vds less than this the JFET acts, not as a constant current generator, but as a resistance.

That was my reasoning behind the boosted supply rail in the second circuit.

(4) For some reason, this type of constant current generator is never very happy and is inclined to be temperamental. A far better performing , but less elegant, constant current generator can be made with a couple of bipolar junction transistors.

Yes, not the best, but wanted to see if it could be done with the absolute minimum of parts, seems maybe not.

(5) The TL431 is a shunt regulator so it cannot source current to drive the base of the series pass power bipolar transistor. Thus the current for the base is the constant current passed by the JFET less 1ma required for the TL431 to operate. Depending on the particular JFET used this would give a base current of 1 to 8 mA. The Hfe (current gain) of the BD139 is 25 so the maximum current the BD139 could output would only be 25mA to 200mA. Because of other factors the output current would be less than this.

Again I had the advantage of being able to measure the hFE of the bd139, prior to use in the circuit. Can't recall the precise figure, but I know that it was over 100. I had only intended for the circuit to work at 100mA or less. According to the graph in the data sheet of current gain vs collector current the gain is around 80-90 within this range, except at very low currents.

(6) Even if a resistor replaced the JFET, this is still a very high gain circuit. The TL431 needs a high gain to achieve its accuracy. This means that frequency tailoring will be required to ensure frequency stability. You can see evidence of this in the TL431 data sheet in the applications section which shows a PSU similar to your #1 PSU but without the JFET constant current generator. The capacitor shown shapes the overall frequency response to ensure stability. This is standard practice for high gain feedback amplifiers. Our old friend the uA741 op amp has a massive 30p capacitor fabricated into the circuit for this sole purpose.

Frequency compensation, now that's definitely a grey area for me, I'm aware of its existence, but not at all clear as to when and where to use it and how to calculate suitable component values for this purpose.

(7) Finally a general point. There is no decoupling on PSU#1. Just to give you a clue, if you look at a well-designed piece of analogue commercial equipment you will often find capacitors splattered all over the place. This is especially so with high end audiophile amplifiers. The reason for this is that to the amplifier the supply lines and other critical points must look like OV (ground) to them, especially as the frequency increases. If that is not the case the physical circuit bears no resemblance to schematic that you work to. This is a big and complicated subject but for the average circuit you can get by with a few basic rules. There is a standard saying with analogue designers: ‘hundred nan to deck’. This means decouple all critical areas with a 100nF low loss capacitor (normally ceramic) to ground or OV.

That was a careless oversight on my part. Iam aware of the importance of decoupling. There was a time when I thought it a pedantic luxury, but I've learned from past difficulties!

That takes care of the high frequencies in most cases, but you also need to decouple at low frequencies too, normally with a 47uF upwards low loss aluminum electrolytic or tantalum capacitor.

Decoupling is not just a thing done by some pedantic analogue design engineers, it is essential. You are probably thinking that you have seen hundreds of circuits with no decoupling. And so you will. But my experience is that many home constructors come to grief because of frequency instability cause by insufficient loop frequency tailoring, no decoupling and poor layout. Decoupling is not always necessary, but it is always a wise precaution. Take the uA741 for example. It is a very friendly opamp and is slugged down to have a narrow frequency response so will tolerate almost anything, but it will still perform best with a good layout and decoupling.

So in summary Gordon you innovative design has been let down by the shortcomings of practical components- they are just not up to your high expectations. I’m sorry to say this but for the reasons already outlined and other reasons which I haven’t gone into, this circuit would be difficult to tame. But no worries, I will post a circuit that will make up for my negative view and give you something to really get your teeth into.

Thanks for taking the time to post such extensive feedback on my circuit. I look forward to seeing its successor!

 

[spec]

Hi Gordon,

Good that you measured the parameters of the actual components. That is an excellent way forward for a one-off design. You obviously know a lot more than I first thought. In view of what you say, it would seem that frequency instability was the overriding problem with PSU #1.

About frequency stability. It is a big subject and I am by no means an expert, but I do know the theory. Instability is caused by two general things: parasitics and unsuitable open loop frequency response. The former is corrected by good layout, screening and decoupling in general. The latter is more complex but it is not difficult to get the basics. You can then move on to more advanced stuff without to much maths. But when you get into control theory, things get heavy. At that point I always called a specialist in.

Fundamental stability criteria for feedback voltage amplifiers.

Negative feedback basically sacrifices gain for predictable and consistent performance. A typical opamp, like the uA741, has a voltage gain of 120dB (x1M) at DC, so if you configure it for a gain of 1 you are chucking away one hell of a load of gain and giving one hell of a load of feedback. The only reason why the opamp does not oscillate is because of the 30pF internal capacitor which starts rolling off the open loop frequency response at 3Hz on a single unit slope. A single unit slope just means an open loop gain falling by -20dB (-x10) per decade of frequency, so that at 30Hz the gain will be down from 120bB (x1M) to 100dB (x100K). At 300Hz the gain will be down to 80dB (x10K) and so on. This is the attenuation rate provided by a single reactive unit, in this case the 30pF capacitor.

To make an oscillator, as you no doubt know, the input and output must be in phase so you get regeneration and endless oscillation. Each capacative unit slope introduces a 90 degree phase lag, so if the gain continued reducing at unit slope thru 0dB there would be no chance of oscillation due to feedback, and you would have a phase margin (safety margin) of 90 degrees. This is exactly the case for a uA741.

If, however, you have another reactive element in the circuit, either intentional or parasitic, you will have a 180 degree phase change so the feedback signal will not be negative but positive and, sure enough, you will have an oscillator. For some reason, the oscillations are typically 4MHz, or higher if MOSFETs are involved. An oscillating audio power amp, for example, may appear to be working OK, but it will sound a bit odd and sooner or later it will take out your speakers and the amps transistors.

There is just one critical fact to remember and you will have the basics of frequency stability nailed. If you draw a graph of open loop frequency response and on the same graph and using the same scale you plot closed loop frequency response, it is the point (intersection) where the two graphs meet that is important. As a general rule, providing the open loop gain intersects at a unit slope (90deg lag) or less the feedback amplifier will be stable. In theory, it does not matter what the open loop frequency response does at frequencies before or after the intersection point, but it is best if the slope at lower frequencies does not exceed a unit slope.

I have pretty bad flue at the moment and as the proposed PSU circuit needs a bit of detailed work, I don't feel up to it. Hopefully I should be better in a couple of days.

spec


uA741 Data Sheet (notice how well behaved the open loop frequency responses shown in figure 7 is- this is one reason why the uA741 reigned supreme for years)
https://www.ti.com/lit/ds/symlink/ua741.pdf

 

[tomizett]

Hi Spec and Gordon. I've nothing technical to contribute here - but it's great to read this kind of interesting, in-depth, technical and polite conversation. It's general enough and sufficiently well-explained that the posts here will be of benefit to people working on all manner of circuits. In particular, Spec, that's a neat description of the Nyquist stability criterion (a lot of stuff fell into place for me the first time I saw that formally explained!).

Thanks all for spending the time on this.

That said, I do have a technical question, but I'll post that in a bit when I've some more time...

 

[Gasboss775]

Yes, thanks very much Spec, that was indeed thorough, yet equally succinct and understandable. I expect it will take me a few repeat reads, to get this very clear in my head such that I may be able to put it into practice, which isn't by any means a criticism of Specs fine description, rather of my poor memory! Given that he has the flu, it is very impressive! I look forward to reading some more of Spec's wisdom.

 

[spec]

Hi Spec and Gordon. I've nothing technical to contribute here - but it's great to read this kind of interesting, in-depth, technical and polite conversation. It's general enough and sufficiently well-explained that the posts here will be of benefit to people working on all manner of circuits. In particular, Spec, that's a neat description of the Nyquist stability criterion (a lot of stuff fell into place for me the first time I saw that formally explained!).

Thanks all for spending the time on this.

That said, I do have a technical question, but I'll post that in a bit when I've some more time...

Yes, thanks very much Spec, that was indeed thorough, yet equally succinct and understandable. I expect it will take me a few repeat reads, to get this very clear in my head such that I may be able to put it into practice, which isn't by any means a criticism of Specs fine description, rather of my poor memory! Given that he has the flu, it is very impressive! I look forward to reading some more of Spec's wisdom.

You are both very kind

 

[Tony Stewart]

You probably had $10.00 in you hand.
.....and you thought you would give her a good tip for cleaning the bed. Good thing you checked out before her boyfriend came back.

One of my staff was in San Paulo Brazil fixing some clean room HDD servo equipment when he joyfully signaled to the lady (also wearing an operating room mask& suit )that he fixed it perfectly **broken link removed**



She turned red beneath her mask and in Brazil that means he wanted...
to do her.

 

[spec]

 

[atferrari]

One of my staff was in San Paulo Brazil fixing some clean room HDD servo equipment when he joyfully signaled to the lady (also wearing an operating room mask& suit )that he fixed it perfectly **broken link removed**



She turned red beneath her mask and in Brazil that means he wanted...
to do her.

Well, not sure about how commonly understood as that (been going to work in Brazil, many times a year for 30 years) but much better is just a "thumb up" sign which works even instead of a smile if fits the occasion...

A more complete gesture, based on the above is frankly obscene, far from just suggestive...

Yes, it is better to be very careful when abroad...! Oh yes.

 

[spec]

​

ERRATA
(1) Frequency stability measures have not been incorporated
(2) Place 0R1 resistors in series with the sources of all MOSFETs
(3) Q18 should be BC546
(4) please don't laugh at the cockle in the trace at the bottom of C14. That is a crime but I just couldn't be bothered to sort it and needed to go to bed

 

[ronsimpson]

If Q16, Q19, Q20, Q23 do not have the exact same gate turn on voltage the current will not share equally.
It would help if each had its own source resistor.

The error amp has no frequency compensation. What keeps it from oscillating?

 

[spec]

If Q16, Q19, Q20, Q23 do not have the exact same gate turn on voltage the current will not share equally.
It would help if each had its own source resistor.

nice one ron

The error amp has no frequency compensation. What keeps it from oscillating?

The OPA192 error amp does have built in frequency compensation. It starts rolling off at 2Hz on a unit slope and bites the dust at 10Mhz. There is a voltage divider in the loop R18/R19 of 5:1 but that is roughly cancelled by the gain of R23 and R17. The dominant pole will be R17 and the total miller capacitance of the MOSFETS. So it is possible that at the closed loop intersection there may be a 180 deg phase change. To answer your question, I suspect nothing much will stop the PSU from oscillating apart from prayers, a not uncommon approach for many PSUs. I intend that the PSU will be adaptable to suit a wide range of opamps.

I have been promising GORDON ROBERTSON a PSU to experiment with in place of the TL431 circuit, but have been unable to get on with it due to flu. In the end I thought I would just do it and be damned. I did know that the frequency stability had not been sorted, but hadn't looked at in any detail.

I was going to do a write-up explaining that the circuit has a very low drop out voltage and can be scaled up and down in terms of output current, output voltage, and precision. The parallel MOSFETS were literally chucked in at the last moment and are only there to dissipate the high power caused by the output voltage being variable. I should really have stuck with a fixed 15V output, and a single MOSFET to simplify things to start with.

Q18 probably looks strange, but is has some benefits which will be covered along the way.

In summary, I think the PSU circuit is OK in principle, but the frequency response probably does need sorting. I haven't checked the details. By the way, if anyone has any ideas about the open the loop voltage gain and how to guarantee a good phase margin, especially you ron, please post. Just one word of caution. The output stage of a PSU is always problematic because, quite simply, you do not know what will be connected to the output. One dodge is to put an inductor in series with the output. If this is wound with really thick wire it would not affect the voltage stability too much. I would like to avoid the inductor approach if possible though, mainly because of size.

I am going back to bed now

 

[Gasboss775]

Thank you very much Spec, that was more than I was expecting. I will look at it in more depth in a bit.
I wasn't expecting you to turn that little circuit of mine into a bench PSU, but it will be interesting to investigate if not quite on the scale implied by your schematic.

I have started one of these jobs you wish you hadnt. I'd had a box of CMOS logic chips stuffed away in a cupboard as I hadn't been doing anything digital for awhile. Iam sorting through them, categorizing them and putting them in little drawers! I started playing with some of the chips to break the monotony. Need to get it finished today.

 

[spec]

Thank you very much Spec, that was more than I was expecting. I will look at it in more depth in a bit.
I wasn't expecting you to turn that little circuit of mine into a bench PSU, but it will be interesting to investigate if not quite on the scale implied by your schematic.

Hi Gordon,

No worries.

I had planned to start with a very simple circuit, illustrating the architecture of the PSU and describing the various sections. Then build up to the circuit posted. But in my present state it proved too much. Instead I hit you with the whole shamola so that you would have something to get your teeth into and just pick the bits you needed for your investigations. For example you only need one voltage source, rather than two, and the current protection (Q17) could be dropped, as could the variable output. If the output current were lowered to 1A from 6A that would mean that the reservoir capacitor value could be divided proportionally as could the mains transformer current capability.

I have started one of these jobs you wish you hadn't. I'd had a box of CMOS logic chips stuffed away in a cupboard as I hadn't been doing anything digital for awhile. I'am sorting through them, categorizing them and putting them in little drawers! I started playing with some of the chips to break the monotony. Need to get it finished today.

Ah yes- sorting listing and categorizing. That is one of the things that electronics people have to suffer, but when things start going wrong and you are surrounded by a mountain of expensive dead transistors, a bit of sorting can be very therapeutic. At least it is something you can do and get results without too much thought